Portfolio Monitoring That Never Sleeps

Tier 1

Cascade Warning

High confidence. Cross-default cascade risk detected across 2+ borrowers with connected exposure.

Routes to: pager credit committee

Response: within 24 hours

Calibrated to fire under 5x per quarter per 200-facility fund.

Tier 2

Storm Forming

Medium confidence. Single-borrower covenant ratio deteriorating, debt-service coverage dropping past warning threshold, unplanned capex mid-quarter.

Routes to: portfolio manager daily digest

Response: within 5 business days

Per-PM sensitivity controls tune the floor.

Tier 3

Watch

Low confidence. Early signals, single data-point drift, not yet indicative of stress. Logged for pattern detection.

Routes to: weekly portfolio summary

Response: no immediate action

Feeds quarterly sensitivity recalibration.

Tier 2 Density Cap

3 Signals Per Day. Hard Cap. No Digest Blindness.

A daily digest that bundles twelve lukewarm signals is just a junk drawer with a calendar. COVENANT caps the Tier 2 daily digest at the 3 most relevant signals per portfolio manager, with a steady-state guarantee of under 5 Tier 2 alerts per week.

When signal volume would exceed the cap, the system makes a hard decision: promote the highest-confidence candidate to Tier 1 (Cascade Warning, pager), or demote the lowest to Tier 3 (Watch, weekly digest). Never overflows.

In the alert preferences UI: "Your current digest is averaging 2.4 signals per day." The PM sees the cap working in real time.

Tier-2 Auto-Promotion Audit Log

Every Promote / Demote Decision Is Logged In The PM's Own Audit Trail

When the density cap forces a promote-or-demote, the system writes a structured audit row that the PM can re-open six months later. Confidence delta, alternatives considered, and a deep link to the underlying signal. Auto-triage logic stops being a black box.

During postmortem, the PM clicks audit_link and sees the full alert tree: which signals were considered, why this one beat them, what the confidence model weighted. If the model was wrong, the row is the receipt.

Conservative-By-Default

Sliders Ship Tuned For The Fiduciary, Not For The Vendor

COVENANT's per-PM sensitivity sliders ship with conservative defaults that catch roughly 95% of correlated breach signals as Tier 1. PMs can dial sensitivity down to fit their portfolio's volatility profile, but the factory default catches the "we should have been paged" cases where demoting a signal would later constitute a breach of fiduciary duty.

Every slider adjustment is logged, timestamped, and reasoned - the PM enters a short note. When the inevitable postmortem happens, the audit log answers "why was this signal demoted" with a dated reason in the PM's own words. The tuning tool is not a vendor liability shield. The defaults are the shield.

COVENANT directs scarce human attention to the 20% that actively requires it, while fully automating the 80% that has clean data pipelines.

Every sub-A borrower ships with an auto-generated Path to A action list: "Re-negotiate reporting requirements in next amendment cycle," "Deploy Onboarding Wizard for this borrower," "Request API integration for upcoming refinancing." Portfolio coverage compounds as the action list closes out.

Parsing Confidence, Not Borrower Default

An F Means The AI Could Not Read The Document. It Does Not Mean The Borrower Defaulted.

The most common silent-failure mode in cloud covenant tools is conflating data quality with parsing confidence. A scanned PDF with a watermark looks identical to a scanned PDF that says "we breached our covenant" - if the parser fails on either, the system goes quiet. COVENANT refuses that conflation. Every F surfaces the precise failure mode and the action it triggers.

An F is a transparent admission that the radar could not see. The PM gets a named failure mode, a routing decision, and a tracked request. The system never goes silent on a borrower it cannot read.

Explainability Dashboard

A Single Glance Tells The Risk Officer Where The Radar Has Coverage

Open the portfolio view. The header is a coverage map: how much of the book is on clean pipelines, how much is delayed, how many borrowers the parser could not read, and how many compliance requests are outstanding right now.

Risk officers stop seeing "the radar found a breach." They start seeing "the radar covers 95% of the book and is actively closing the remaining 5%." The blind spot becomes a tracked metric, not a hidden liability.

A-F Grading: Active Enforcement

True Risk Management Is Defined By How The System Behaves In The Dark

When a borrower hits a D or F grade, COVENANT does not wait. The system auto-generates a templated compliance-enforcement request to the borrower's CFO or CRO, referencing the specific contractual reporting obligations the borrower signed, requesting the missing data, and opening a tracked response timer.

Non-responsive borrowers surface as a daily Compliance Friction report on the PM dashboard, sorted by days-since-request and by underlying credit exposure. The 20% is not celebrated as a blind spot. The 20% is hunted until it becomes the 80%.

The companies with the worst financial-reporting hygiene are historically the first to default in a liquidity crisis. Active enforcement does not pretend otherwise; it goes and gets the data.

CISO Fast-Track Bridge

The Translation Layer, Not A Substitute For Institutional Audit

The COVENANT SOC 2 Equivalence Document is a CISO Fast-Track Bridge for early adopters: it translates the local-inference architecture into the vocabulary a cloud-era SOC 2 checklist already understands. It is not a substitute for an independent audit. It is the bridge that gets early-mover credit funds deployed while the independent audit completes.

Because borrower financial data never leaves the fund's environment, the standard SaaS security risks are bypassed entirely. The architecture itself is the primary security control. The Bridge is the roadmap that proves the local architecture requires a lower burden of proof.

Independent Big-4 SOC 2 Type 2 audit target: Q2 2027. Funded by founder-tier subscription revenue. Published on a public calendar so CISOs can plan around it rather than squint at a self-attested document.

Request The Fast-Track Bridge →

SOC 2 Equivalence v2: Fortress Framing

We Are Not Safer Because We Passed The Checklist. We Are Safer Because We Don't Have The Cloud Problem.

SOC 2 was written for a world where the vendor holds your data. Every control on that checklist exists to govern how the vendor protects what they hold. COVENANT collapses the threat model by never holding the data in the first place. The structural answer is mathematically stronger than the checklist answer.

"Decentralizing the honeypot is mathematically safer than protecting a central target." A SOC 2 audit certifies the central target is well-protected. COVENANT eliminates the target.

Hardware EOL Strategy

"What If Apple Kills The M-Series?" Has A Documented Answer.

Apple Silicon is the current reference platform because unified memory and the on-package Neural Engine make this latency profile possible on a desk-side appliance. It is not the only platform COVENANT can run on. The application is bound to MLX, Core ML, and Metal - portable across every M-series generation Apple has shipped - and the inference backend is abstracted so the same data pipeline, parser, alert engine, and UI run unchanged on Snapdragon X Elite, AMD Strix Halo, or NVIDIA workstation hardware if the M-series is ever discontinued.

M4 → M5 migration is documented as a same-week appliance swap. Hardware refresh is voluntary unless Apple drops macOS security patches for the deployed generation - typically a 6-to-8-year window. The fund chooses its own refresh calendar.

Full roadmap, generation-to-generation migration runbook, and 6-year TCO model are published in the COVENANT Hardware EOL Strategy. Available on request alongside the CISO Fast-Track Bridge.